Companies Tightening AI Access as Costs Surge
In various sectors like technology, banking, and entertainment, organizations are significantly curbing employee access to AI tools due to skyrocketing costs, which have soared into millions, according to internal documents from several companies.
Recent reports indicate that firms such as Atlassian, Adobe, Amazon, Citi, and GitHub are grappling with a financial strain as their AI expenditures have, in some instances, tripled. This has prompted them to restrict access to advanced AI models and urge employees to limit their usage.
This predicament seems to stem from two key factors: companies adopting AI rapidly without grasping the financial implications, coupled with AI providers shifting from fixed-rate subscriptions to pay-per-use pricing based on AI tokens utilized in each interaction.
Citi has taken particularly bold measures, entirely blocking access to the latest Claude and ChatGPT models. An internal email revealed that the bank disabled Claude Opus 4.6 and 4.7 as well as GPT-5.5 on June 24, highlighting that these models “consume significantly more AI credits per interaction.”
Before these restrictions, Citi advised employees to use lower-performance models unless absolutely necessary. The communication clarified that because AI tokens are pooled across the organization, those with more intensive AI tasks could deplete shared resources more rapidly. Recommendations were given on which models to use for different tasks, suggesting GPT-5.3-Codex for basic questions and code generation while reserving advanced options like Claude Sonnet 4.6 for complex reasoning.
These changes at Citi were made in response to a shift from flat subscription pricing to a pay-as-you-go model in June. Currently, the bank is monitoring usage patterns closely to catch overspending early and has introduced budget controls. However, when contacted, Citi contested this portrayal, stating that they had not disabled any models or set token limits, despite internal documents suggesting otherwise.
Atlassian has faced notable cost increases as well. The well-known software company, which offers the Jira development tool, has ended unrestricted access to AI and set up an internal dashboard for employees to monitor costs. This dashboard indicates the company expects to spend over $120 million on AI tools in the fiscal year, with costs projected to rise from $5 million in August 2025 to upwards of $15 million by May 2026. Atlassian contended that these figures didn’t accurately capture their AI usage but did not clarify which numbers were wrong.
An Atlassian employee noted mounting frustration among staff members who have reshaped their work habits around AI. “We’ve seen a lot of complaints that even though people have revamped their workflows to optimize AI, they run out of resources quickly, especially when using agents or modern cloud models,” the employee said. “Slack is full of messages asking, ‘What should I do now?’
Breitbart News had previously reported an unnamed company, likely Amazon, inadvertently spending $500 million on Anthropic’s Claude AI in just one month.
While such a figure stands out, other recent occurrences show how quickly AI expenses can escalate. In April, Google Cloud clients were surprised by an unexpected $18,000 bill after a security breach, despite budgeting only $7. Earlier in May, the creators of OpenClaw disclosed that OpenAI API tokens had consumed $1.3 million within a single month.
According to Axios, companies that were once keen to invest in AI are now facing substantial costs without guaranteed returns. This trend has become so common that it has even been dubbed “token maxing,” which refers to the practice of maximizing AI token usage for sometimes questionable purposes.
The discourse surrounding AI is becoming increasingly divisive within the business realm as its transition from novelty to practical applications unfolds. Wynton Hall, social media director at Breitbart News, discusses the need for the MAGA movement to articulate a position on AI that serves humanity without yielding control to Silicon Valley or allowing Chinese dominance.
